Cross-Regional Learning for Sparse-View CBCT Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CBCT reconstruction methods face challenges in sparse-view scenarios due to high computational cost, poor performance on anatomies with complicated structures, and inadequate utilization of cross-view relationships, leading to suboptimal image quality and radiation dose concerns.
Innovation Solution
A novel sparse-view CBCT reconstruction framework, C2RV, leverages cross-regional and cross-view feature learning through multi-scale 3D volumetric representations and scale-view cross-attention to enhance point-wise representation, aggregating multi-view pixel-aligned and multi-scale voxel-aligned features for improved attenuation coefficient estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If the number of projections is reduced to reduce radiation dose, then radiation dose is reduced, but image quality deteriorates
Solution Approach 1:
The patent transforms the reconstruction problem from 2D slice-by-slice reconstruction to 3D volumetric reconstruction by introducing multi-scale 3D volumetric representations. This dimensional change enables the model to leverage spatial correlations across the entire 3D volume, improving reconstruction quality from sparse projections while maintaining reduced radiation dose.
Solution Approach 2:
The patent segments the feature extraction process into multiple scales by generating multi-scale 3D volumetric representations at different resolutions. This segmentation allows the model to capture both global structural information and local detailed features, thereby improving image quality reconstruction from sparse views without requiring increased radiation exposure.
2Reliability
If conventional CT reconstruction methods are extended to CBCT, then reconstruction capability is achieved, but computational cost increases
Solution Approach 1:
The patent segments the feature processing into multi-scale representations, where coarse-scale features capture global structures and fine-scale features capture local details. This segmentation reduces the computational burden compared to processing full-resolution 3D data, while maintaining reliable reconstruction capability for CBCT.
Solution Approach 2:
By formulating the problem as 3D volumetric reconstruction rather than 2D slice reconstruction, the patent achieves better reliability for CBCT while optimizing computational efficiency through the hierarchical multi-scale approach that reduces overall computational complexity.
3Ease of operation
If all projection views are processed equally, then processing simplicity is maintained, but feature utilization efficiency deteriorates
Solution Approach 1:
The patent applies local quality by allowing different projection views to be processed with view-specific attention weights. Instead of treating all views equally, the model learns to assign different importance levels to different views based on their informational content, thereby improving feature utilization efficiency while maintaining reasonable processing complexity through the attention mechanism.
4Ease of operation
If single-scale reconstruction is used, then processing simplicity is maintained, but reconstruction quality on complex anatomies deteriorates
Solution Approach 1:
The patent segments the reconstruction into multiple scales, where each scale captures different levels of anatomical detail. This segmentation enables the model to reconstruct complex anatomies with both global structural accuracy and local detailed fidelity, significantly improving reconstruction quality compared to single-scale methods while managing processing complexity through hierarchical processing.
Solution Approach 2:
By adding the scale dimension to the reconstruction process, the patent enables simultaneous capture of both coarse and fine anatomical features. This multi-scale approach dramatically improves reconstruction quality on complex anatomies while the hierarchical structure keeps processing manageable.
Data Source
AI summary
A cross-regional and cross-view learning (C2RV) framework is provided for sparse-view reconstruction in cone-beam computed tomography (CBCT) by advantageously leveraging cross-region and cross-view feature learning to enhance representation of a point in 3D space before estimating an attenuation coefficient of the point. Specifically, multi-scale 3D volumetric representations (MS-3DV) are first introduced, where features are obtained by back-projecting multi-view features at different scales to the 3D space. Explicit MS-3DV enable cross-regional learning in the 3D space, providing richer information that helps better identify different internal anatomy structures. Hence, features of the point can be queried in a hybrid way, i.e. multi-scale voxel-aligned features from MS-3DV and multi-view pixel-aligned features from projections. Instead of considering queried features equally, scale-view cross-attention (SVC-Att) is used to adaptively learn aggregation weights by self-attention and cross-attention. Finally, multi-scale and multi-view features are aggregated to estimate the attenuation coefficient.